deprecate sharded_ddp training argument

[ji-huazhong]

What does this PR do?

This PR deprecates the sharded_ddp training arguments, since ShardedDDP has been upstreamded to PyTorch, so users can use the fsdp traning parameter instead.

---

According to fairscale(see), PyTorch FSDP is the recommended method for scaling to large NN models. I think Sharded-DDP is dead and it's time to say goodbye to this library.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[sgugger]

That makes sense thanks! Note that while removing it entirely from the doc is fine, we can't abruptly remove it from the library like this. We will need to properly deprecate it first and in two-tree minor versions we can fully remove it.

[ji-huazhong]

That makes sense thanks! Note that while removing it entirely from the doc is fine, we can't abruptly remove it from the library like this. We will need to properly deprecate it first and in two-tree minor versions we can fully remove it.

@sgugger Thanks for pointing this out. I have rolled back the code changes and added a warning that sharded_ddp will be deprecated. Could you take a second look at this PR?

[sgugger]

Thanks for iterating!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.